Add a concept of 'machine_type_spec'.

In Kubernetes (and Marathon which is a framework for Mesos) we don't have flavors. Passing a flavor using `--machine_type` flag doesn't make sense. We do have a possibility of specifying resources limits though (like memory and cpu). I'd like to create a common flag for these resources which will be used by some of the providers (like Kubernetes and Mesos). I would like to know your opinion before I start the implementation. Here are the solutions I see:
- leave current `--machine_type` flag untouched and introduce a second flag: `--machine_type_spec`. Resource limits would be passed in the following way: `--machine_type_spec=|cpu:2|,|memory:1024|` - we would have to agree on the format. I have seen that you already use this format in the JSON with benchmarks results but I'm open for new ideas.
- similar to the previous one, but resources limits would be passed using current `--machine_type` flag, i.e. `--machine_type=|cpu:2|,|memory:1024|`
- introduce new flag for each resource type, i.e. `--cpus=2`, `--memory=1024`.

Personally I would stick to the first or third option. Please let me know what you think about this. Thanks!